Notice This manuscript is a non-peer reviewed preprint submitted to EarthArXiv. It has been submitted for publication to GJI on 09/10/2021 with reference ID #GJI-21-0923. Newer versions may be moderately different with slight variations in content. Manuscript details Title: Kathmandu Basin as a local modulator of seismic waves: 2D modelling of nonlinear site response under obliquely incident waves

The 2015 Mw 7.8 Gorkha, Nepal earthquake is the largest event to have struck the capital city of Kathmandu in recent times. One of its surprising features was the frequency content of the recorded ground motion, exhibiting a notable amplification at low frequencies (< 2 Hz) and a contrasting depletion at higher frequencies. The latter has been partially attributed to the damper behaviour of the Kathmandu basin. While such weak high-frequency ground motion helped avoiding severe damage in the city, the catastrophic outcomes of earlier earthquakes in the region attest to a contrasting role of the Kathmandu basin as a broadband amplifier, in addition to possible source effects. Given the possibility of future strong events in the region, our main objective is to elucidate the seismic behaviour of the Kathmandu basin by focusing on site effects. We numerically model 2D P-SV wave propagation in a broad frequency band (up to 10 Hz), incorporating the most recent data for the Kathmandu basin geometry, soil stratigraphy and geotechnical soil properties, and accounting for the non-linear effect of multi-dimensional soil plasticity on wave propagation. We find that: 1) the Kathmandu basin generally amplifies low frequency ground motion (< 2 Hz); 2) waves with large incidence angles relative to vertical can dramatically amplify the high frequency ground motion with respect to bedrock despite the damping effect of soil nonlinearity; 3) the spatial distribution of peak ground motion amplitudes along the basin is highly sensitive to soil nonlinearity and wave incidence (angle and direction), favoring larger values near the basin edges located closer to the source, as observed during the 2015 event. Our modelling approach and findings can support the ongoing resilience practices in Nepal and can guide future seismic hazard assessment studies for other sites that feature similar complexities in basin geometry, soil stratigraphy and dynamic soil behaviour.

ley were below the estimations of ground motion prediction equations (GMPE) at frequencies here focus on site effects: despite possibly short source-to-site distance, we ignore complexities 93 arising from fault finiteness by limiting our study to the assumption of plane wave incidence. by active faults and is not symmetrical, we also investigated the sensitivity of the ground motion 119 inside the Kathmandu basin to the obliquity of incident waves.

120
In the following, we first present the studied area, and the methods and data used for numer-121 ical modelling. Then, we report our results on site effects in the Kathmandu basin at low and basin. Last, we summarise our main findings and perspectives for future research. Here we study one of its 2D cross-sections that extends in the east-west direction.

127
We first created the 3D geometry of the Kathmandu basin by combining the sub-surface images

133
We set three sediment layers for the basin and consider that the shallowest layer is nonlinear.  Table 1. To simplify 142 the evaluation in the following, we virtually divided the basin into three sections, referred to 143 hereafter as the western, middle, and eastern parts of the basin, respectively, as denoted by I, II, 144 and III in Fig. 1b.

145
In the absence of detailed knowledge of soil nonlinearity properties, we assumed that only 146 the first layer is nonlinear, given its soil type, relatively shallow depth and low velocity. We set   Pronounced low-frequency ground motion in the Kathmandu Basin also occurs under oblique 208 wave incidence. In Figure 5, we present the soil-to-rock spectral ratios for three wave incidence 209 angles relative to the vertical axis: 30 degrees from west, 0 degrees and 30 degrees from east.

210
An incidence of 30 degrees is plausible for the regional seismotectonics and useful for com-211 parison purposes. We used the Northridge input and considered soil nonlinearity in all the three 212 cases. The change of incidence angle causes local variations in the fundamental frequencies and 213 spatial pattern of the spectral ratios. If incidence is from west (east), the largest amplification 214 appears in the western (eastern) side of the basin. In all cases, the largest soil-to-rock spec- in the Northridge case is seen at TVU by slightly smaller spectral ratios above ∼1 Hz. In ad-230 dition, both nonlinear cases produce a slight shift in the resonance frequencies (from 0.5 Hz On the other hand, despite basin nonlinearity, a critically oblique wave incidence can boost 252 the high frequency ground motion (> 2 Hz) inside the Kathmandu basin with respect to the 253 outer rock. We performed an additional set of simulations with gradually increased incidence 254 angles and adopting the Northridge input. Figure 7a shows the soil-to-rock spectral ratios in 255 linear simulations with wave incidence angles of 30, 40 and 45 degrees from East. The basin 256 strongly amplifies ground motion over a broader frequency band at increasing incidence angle.

257
At 40 degrees of incidence, the amplification above 1 Hz is concentrated at the edges of the 258 three sections of the basin, and the soil-to-rock spectral ratio reaches a factor of ∼10 below 259 5 Hz. At 45 degrees of incidence, the amplification is dramatically larger all over the basin. The 260 theoretical value of refraction due to impedance contrast (by Snell's law) ranges between 20 261 and 28 degrees for the 1D simplification of the soil strata. Our additional 2D simulations prob-262 ing more incidence angles (supporting figures in SI) show that strong broadband amplification 263 above 2 Hz occurs at incidences higher than ∼42 degrees. Figure 7b shows the same compar-264 ison but including soil nonlinearity. The soil nonlinearity attenuates the ground motion for all 265 the cases of wave incidence angle. Despite that, the enhanced high-frequency amplification at 266 increasing incidence angle prevails. Such a dominant amplification effect is also seen at the lo-267 cations of two soil stations, TVU and THM (Figure 8): their spectral ratios are a factor of ∼ 5 268 larger at 45 degrees incidence than at 30 degrees incidence, at frequencies > 0.5 Hz.

269
We propose that the incidence angle effect may have contributed to the differences in the    The damping effect of soil nonlinearity in the Kathmandu basin enhances the contrast of peak 293 ground motion amplitudes between the edges and deeper parts of the basin. We analysed the 294 spatial variation of ground motion amplitudes across the basin and how it relates to the basin 295 nonlinearity. Figure 9 displays the comparison of PGA along the basin length between linear 296 and nonlinear cases, together with the maximum -total-strain reached in the nonlinear layer.

297
Results are shown for the two input cases: pulse-like (top) and Northridge (bottom). Wave 298 incidence is vertical in both cases. In the pulse-like input case, the shallower parts close to higher everywhere. The stress-strain curves at locations close to eastern and western edges of 301 the basin, Figure 9 (b, d), show higher complexity of the loading cycle for the Northridge input, 302 consistently with its larger number of zero-crossings (See the discussion in 2). For both input 303 motions, in the linear simulations, the PGA values are comparable all along the basin length, 304 although the combined effects of basin geometry and soil stratigraphy lead to slightly larger 305 PGA values close to corners and section boundaries (e.g., at x=13, 15, 18.5, 22, and 30 km). In 306 the simulations with soil nonlinearity, PGA is strongly reduced everywhere there is a sufficiently 307 thick nonlinear layer below but remains high elsewhere (details in SI). The local peaks in the 308 deeper parts of the basin mostly disappear, and the PGA shows notable contrasts near the edges 309 of basin sections favouring larger amplitudes where nonlinear soil is not thick. Despite higher 310 level of nonlinearity triggered in such thin layers (e.g., x=22 km), the PGA in the proximity 311 remains large, such that the PGA ratio between basin corners and deeper sections can rise to a 312 factor of 5, as seen in the case of Northridge input (at x=22 km vs x=25 km).

313
The direction of wave incidence can cause further variation of triggered basin nonlinearity. 314 Figure 10 compares the basin response to wave incidence from east and west, for the Northridge 315 input. The incidence angle equals 30 degrees in both cases. Incidence from east results in larger 316 strains in the eastern section. The effect of such higher nonlinearity on ground motion is rather 317 slight, manifesting as further local variation of PGA in that section. Incidence from west triggers 318 a similar effect on the western section.

319
Given that Kathmandu is inhabited by a dense population and hosting highly vulnerable 320 constructions, our findings of the local variation of the ground motion due to the direction 321 of wave incidence and soil nonlinearity warrant further research on regional seismic hazard 322 including these factors. Our study is limited to plane wave incidence, and further investigation 323 of the spatial variability of ground motion deserves a closer look into possible effects of source 324 finiteness and rupture directivity. strain values for the nonlinear layer and set zero strain elsewhere in the 2D plots. Selected locations are denoted by triangles in the 2D plots. Wave incidence is vertical in both cases.

326
We found that the Kathmandu basin typically enhances low-frequency ground motion (< 2 Hz) 327 with and without nonlinear soil behaviour, and regardless of wave incidence angle. This find-328 ing supports and expands the insights from past studies of ground motions produced by the 329 2015 Gorkha earthquake. Here, accounting for the 2D basin geometry, soil stratigraphy and 330 multi-dimensional soil nonlinearity, thanks to the most recent geotechnical data, we find that 331 low-frequency ground motion amplification in Kathmandu should be expected during future 332 earthquakes. 333 We also found that the angle of wave incidence can tremendously boost the high-frequency 334 ground motion across an entire basin, compared to bedrock, despite the damping effect of soil 335 nonlinearity. In our models, ground motion amplification appears prominently (up to a factor 336 of 5) at wave incidence angles larger than ∼42 degrees relative to vertical. We propose that  Figure 10. The effect of basin nonlinearity on peak ground motion for different wave incidence direction.
(a) The comparison of PGA variation along the basin length between linear and nonlinear cases (top), and maximum strain distribution in the basin (bottom), and stress-strain curves at selected basin locations, for the case of wave incidence from west, (b) same as (a) for the case of wave incidence from east. We only show the max. strain values for the nonlinear layer and set zero strain elsewhere in the 2D plots.
Selected locations are denoted by triangles in the 2D plots. The incidence angle equals 30 degrees in both cases. the position of the source relative to the basin, through the effective wave incidence angle, 338 may have contributed to the differences in damage impact between the Gorkha event and earlier 339 earthquakes. Investigating in broadband to what extent such wave incidence effects prevail when